Senior Principal Model Bring-up Engineer

SambaNova; San Jose (CA); On-site; USD 208,000 - 312,000; Full time

Benefits offered by this job

Health insurance

HSA

Dental

Vision

Wellbeing benefits

Job summary

SambaNova Systems, based in San Jose, CA, is seeking a senior ML systems engineer to bring up open-weight LLMs on our RDU accelerators. You will adapt models, map across chips and hosts, compile, and validate numerically against references, then deploy production serving with latency and throughput targets.

You will build shared inference features such as continuous batching, prompt caching, and speculative decoding (FP8/FP4), and contribute to MLIR/XLA/Triton based compilers and tooling.

Qualifications

• B.S. in Computer Science, Electrical Engineering, or related field.
• 8+ years of industry experience in ML systems, inference, or compiler engineering.
• Strong Python and PyTorch internals experience, plus some C++.
• Deep transformer-inference knowledge, including KV caching, GQA and sliding window attention, MoE routing, prefill vs decode, batching, and speculative decoding.
• Hands‑on experience with accelerator bring‑up, ML compilers (MLIR, XLA, TVM, Triton, torch‑mlir), or inference engines (vLLM, SGLang, TensorRT‑LLM).
• Rigor in numerical debugging.
• Comfort with Bazel, Slurm, and large monorepos.
• Clear writing and evidence‑backed claims.

Responsibilities

• Bring up new LLM architectures end to end: model adaptation, checkpoint conversion, multi-chip / multi-host mapping, compile, strict numerical validation, production serving with latency / throughput numbers.
• Build inference features once as shared modules: continuous batching over a paged KV pool, prompt caching, chunked and incremental prefill, dynamic tensor shapes, speculative decoding (EAGLE-3, drafters, MTP), FP8 / FP4, tensor and expert parallelism.
• Debug where model, compiler and hardware meet; land each fix in the right shared component.

Skills

Python

PyTorch internals

C++

Transformer inference

ML compilers

Numerical debugging

Bazel

Slurm

Technical writing

GQA attention

Education

B.S. in Computer Science / Electrical Engineering

M.S. or Ph.D. in related field

Tools

MLIR

XLA

TVM

Triton

torch-mlir

TensorRT-LLM

Job description

SambaNova is a leader in next-generation AI infrastructure, delivering a full-stack inference platform for customers worldwide. At the core of SambaNova's technology is the RDU (Reconfigurable Dataflow Unit) — a chip built on a dataflow architecture rather than the traditional GPU model. Its decode performance is especially strong for agentic workloads like multi-turn agents, code generation, and long-running applications. RDUs are packaged into SambaRack, rack-scale hardware that lets customers deploy state-of-the‑art models with better performance, greater energy efficiency, and faster time to value.

About The Team

We take frontier open-weight models (dense, mixture-of-experts with hundreds of experts, hybrid Mamba/attention, sliding-window attention, multi-token prediction) from Hugging Face to production serving on SambaNova's RDU dataflow accelerators. That means compiling them, proving them numerically correct against the reference, serving them with the full feature set, and merging the work into a shared stack so the next model is cheaper.

About The Role

You will bring new open-weight LLMs, including dense, mixture-of-experts, and hybrid Mamba/attention architectures, up on SambaNova’s RDU dataflow accelerators and into production serving. That means adapting the model, mapping it across chips and hosts, compiling it, and validating it numerically against the reference. You also build inference features like continuous batching, speculative decoding, and FP8/FP4 as shared modules, so each new model costs less to bring up.

Responsibilities

• Bring up new LLM architectures end to end: model adaptation, checkpoint conversion, multi-chip / multi-host mapping, compile, strict numerical validation, production serving with latency / throughput numbers.
• Build inference features once as shared modules: continuous batching over a paged KV pool, prompt caching, chunked and incremental prefill, dynamic tensor shapes, speculative decoding (EAGLE-3, drafters, MTP), FP8 / FP4, tensor and expert parallelism.
• Debug where model, compiler and hardware meet; land each fix in the right shared component.

Required Qualifications

• B.S. in Computer Science, Electrical Engineering, or related field
• 8+ years of industry experience in ML systems, inference, or compiler engineering
• Strong Python and PyTorch internals experience, plus some C++
• Deep transformer-inference knowledge, including KV caching, GQA and sliding window attention, MoE routing, prefill vs decode, batching, and speculative decoding
• Hands‑on experience with accelerator bring‑up, ML compilers (MLIR, XLA, TVM, Triton, torch‑mlir), or inference engines (vLLM, SGLang, TensorRT‑LLM)
• Rigor in numerical debugging
• Comfort with Bazel, Slurm, and large monorepos
• Clear writing and evidence‑backed claims

Preferred Qualifications

• M.S. or Ph.D. in Computer Science, Electrical Engineering, or related field
• Experience with dataflow or spatial compilers
• Experience with kernel DSLs
• Experience with FP8 and FP4 inference
• Experience with Mamba and SSMs
• Experience with expert parallelism
• Experience with multi‑agent AI engineering workflows
• Open‑source inference or compiler contributions

Success in six months
Own a model family from first compile to served and merged; turn a one‑model fix into a shared module others adopt; measurably shorten one long loop for the team.

Base Salary Range:

Base Pay Range

$208,000—$312,000 USD

EEO Policy

SambaNova Systems is an Equal Opportunity/Affirmative Action Employer. All qualified applicants will receive consideration for employment without regard basis of age (40 and over), color, disability, gender identity, genetic information, marital status, military or veteran status, national origin/ancestry, race, religion, creed, sex (including pregnancy, childbirth, breastfeeding), sexual orientation, and any other applicable status protected by federal, state, or local laws.

Benefits Summary For US-Based, Full-Time Employment Positions

SambaNova offers a competitive total rewards package, including the base salary, plus equity and benefits. We cover 95% premium coverage for employee medical insurance, and 77% premium coverage for dependents and offer a Health Savings Account (HSA) with employer contribution. We also offer Dental, Vision, Short/Long term Disability, Basic Life, Voluntary Life, and AD&D insurance plans in addition to Flexible Spending Account (FSA) options like Health Care, Limited Purpose, and Dependent Care. Our library of well‑being benefits available to you and your dependents includes a full subscription to Headspace, Gympass+ membership with access to physical gyms, One Medical membership, counseling services with an Employee Assistance Program, and much more.